Identification of cell signaling pathways based on biochemical reaction kinetics repositories

• Ano de defesa: 2021
• Autor(a) principal: Matos, Gustavo Estrela de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Cell signaling pathways; Equações diferenciais ordinárias; Feature selection; Model selection; Ordinary differential equations; Seleção de características; Seleção de modelos; Vias de sinalização celular
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/45/45134/tde-08032021-211926/

Cell signaling pathways are composed of a set of biochemical reactions that are associated with signal transmission within the cell and its surroundings. From a computational perspective, those pathways are identified through statistical analyses on results from biological assays, in which involved chemical species are quantified. However, once generally it is measured only a few time points for a fraction of the chemical species, to effectively tackle this problem it is required to design and simulate functional dynamic models. Recently, a method was introduced to design functional models, which is based on systematic modifications of an initial model through the inclusion of biochemical reactions, which in turn were obtained from the interactome repository KEGG. Nevertheless, this method presents some shortcomings that impair the estimated model; among them are the incompleteness of the information extracted from KEGG, the absence of rate constants, the usage of sub-optimal search algorithms and an unsatisfactory overfitting penalization. In this work, we propose a new methodology for identification of cell signaling pathways, based on the aforementioned method, with modifications on the cost function that aims to solve the unsatisfactory overfitting penalization. To this end, we use a cost function based on the marginal likelihood of a model producing the observed experimental data. We show how this new cost function automatically penalize complex models, since marginal likelihood approaches tend to select models with intermediate complexity. The new methodology was tested on artificial instances of model selection; for one of the experiments, we solved the model selection problem as a feature selection problem, walking on the space of solutions to get a glance of the surface induced by the defined cost function. With this work, we expect to contribute towards the solution of the cell signaling pathway identification problem, by providing the implementation of a cost function that can be used in a feature selection approach.